Parallel version of a simple sum is 10 times slower

[golroth]

Hello,

I'm currently trying to learn how to use Intel TBB and I tried to write a simple short program using parralel_reduce.

However, the parallel version is 10 times slower than the sequantial one...

I really can't understand why the parallel version is that slower, could you help me figure out why please ?

Here are both of my programs :

parallel_sum_1_over_n.cc

sum_1_over_n.cc

I compiled them with g++ under Fedora 12 linux and my computer has a dual core processor.

Thank you for your help.

[Dmitry_Vyukov]

And what version of TBB?

[RafSchietekat]

"And what version of TBB?"
Dmitriy probably means (sorry...) that before TBB 2.2 the default partitioner was simple_partitioner, and to get acceptable performance with that you have to set an appropriate grainsize (third parameter of blocked_range, e.g., 1000), or otherwise you must explicitly specify auto_partitioner as the thrd argument of parallel_reduce() (the default since TBB 2.2).

[golroth]

Indeed, I am using the 2.1 version and I did not specify neither a partitioner, nor a grainsize.

It works very well now :)

Thank you very much for your help :D